CN112776673B - Real-time energy optimization management system for intelligent networked fuel cell vehicles - Google Patents

Abstract

An intelligent network-connected fuel cell automobile real-time energy optimization management system belongs to the field of fuel cell automobile optimization control. The invention aims to provide a layered real-time energy rolling optimization control intelligent networking fuel cell automobile real-time energy optimization management system for a fuel cell automobile. The invention designs a macroscopic long-time-domain average traffic flow velocity trajectory prediction module, designs a microscopic short-time-domain vehicle speed prediction module, establishes a fuel cell vehicle power system model facing energy optimization control, establishes an energy optimization management problem, designs an upper-layer trajectory rolling optimization controller by using long-time-domain preview information, designs a lower-layer energy rolling optimization controller by using short-time-domain preview information, and transmits a solved control input sequence signal to a power execution control unit of a fuel cell vehicle. The invention excavates the energy-saving space of the fuel cell automobile in the intelligent network traffic environment and obviously improves the fuel economy of the fuel cell automobile in the intelligent network environment.

Description

Real-time energy optimization management system for intelligent networked fuel cell vehicles

technical field

The invention belongs to the field of optimal control of fuel cell vehicles.

Background technique

With the continuous increase in the number of automobiles in our country, the problem of energy consumption and environmental pollution is becoming more and more serious. Energy saving and environmental protection have become the core issues of our country. Fuel cell vehicle is a new type of clean energy vehicle that is very friendly to the environment. Compared with traditional internal combustion engine vehicles, the working process of fuel cells is not limited by the Carnot cycle, the energy conversion efficiency is high, and the terminal emission product is water, which can achieve zero emissions and zero pollution. The large-scale application of hydrogen energy and fuel cell vehicles will be realized, and the number of fuel cell vehicles will reach about 1 million; at the same time, the core key technologies of fuel cells will be fully mastered, and a complete industrial chain of preparation and production of fuel cell materials, components and systems will be established.

Fuel cell vehicles are powered by two power sources, fuel cells and power batteries. Fuel cell is the core energy source of fuel cell hybrid power system. It has high power density and plays a major role in ensuring the normal operation of the vehicle, but its dynamic response speed is slow and it is difficult to adapt to frequent changes in load; power battery It is an auxiliary energy source for the fuel cell hybrid power system. It has a low power density, but has a fast dynamic response speed, which can better cope with large load changes and provide part of the energy that the fuel cell cannot provide when the vehicle is running. . At the same time, when the car is in the braking state, the motor working mode of the car will change from the motor mode to the generator mode, and the power battery can recover the regenerative braking energy generated when the car brakes to charge itself, which can be used to a certain extent. Avoid wasting energy. The two power sources work together to make up for the shortcomings of the slow response speed of the fuel cell and the low power density of the power battery.

The energy flow direction of the fuel cell vehicle is complex, and its working efficiency shows a trend of first rising and then decreasing with the increase of power, which is not a simple linear relationship. In the fuel cell vehicle system, an energy management strategy is required to reasonably allocate the power sources of each power source. power, so that the fuel cell works in the high-efficiency range as much as possible, and the power battery makes full use of the recovered braking energy. Energy management strategy is the core of improving fuel economy, so designing an excellent energy management system is of great significance for fuel cell vehicles. In addition, with the rapid development of artificial intelligence, sensor technology and cloud computing, intelligent networked vehicles are the inevitable trend of future automobile development. Intelligent networked vehicles use their own sensors to sense the surrounding environment, and communicate with other vehicles through communication equipment. Get multi-source driving preview information. Using these multi-source preview information, the vehicle control system can predictably optimize the energy management of fuel cell vehicles in combination with the reference information given by cloud computing, which can provide more room for improvement in the fuel economy of the vehicle.

Patent CN108944900A discloses a fuel cell vehicle energy management control method. The invention formulates a power distribution strategy for fuel cell and power battery according to the current driving demand power of the vehicle and combined with the state of charge of the power battery. However, this invention is a rule-based instantaneous energy management strategy, which can only make decisions on energy distribution according to the current vehicle state, and cannot optimize the working efficiency of fuel cells and power batteries. In addition, the invention does not involve the vehicle energy management strategy in the intelligent network environment, and the energy saving potential of the vehicle has not been fully tapped.

The patent CN109795373A discloses a durability-based fuel cell commercial vehicle energy management control method. The invention adopts a fuzzy logic strategy to distribute the output power of the fuel cell and the power battery, which better improves the fuel economy of the fuel cell vehicle. However, this strategy needs to design the corresponding power distribution fuzzy logic under the condition of known fixed driving conditions, and can only be calculated offline, which has great application limitations in the face of unknown and complex traffic environment with high uncertainty sex.

The patent CN110696815A discloses a predictive energy management method for a networked hybrid vehicle. The invention takes the gasoline-electric hybrid vehicle as the research object, and distributes the energy of the engine and the power battery by introducing the network information. However, the invention does not involve the energy optimization management strategy for fuel cell vehicles. The fuel cell is different from the engine and cannot be started and stopped frequently during operation, and has more stringent constraints. Compared with the hybrid electric vehicle, the fuel cell vehicle has Energy management design is more difficult.

To sum up, although the currently published patents have involved some solutions for energy management of fuel cell vehicles, they are basically rule-based methods, which make it difficult to obtain optimal fuel economy. Economical, the algorithm to be used requires a large amount of calculation, and the complete working condition information needs to be known in advance, and it can only be calculated offline. In order to solve the contradiction between the real-time performance of the optimization algorithm and the energy-saving effect, and to further tap the energy-saving potential of fuel cell vehicles in the environment of intelligent networked transportation, designing a real-time energy optimization management system for intelligent networked fuel cell vehicles is an urgent problem to be solved and full of challenges.

SUMMARY OF THE INVENTION

The purpose of the invention is to use historical vehicle speed sequence to predict vehicle speed information in short time domain, and combine the vehicle speed preview information in long time domain and multi-power source system model to propose a layered real-time energy rolling optimization control for fuel cell vehicles. Real-time energy optimization management system for battery vehicles.

The steps of the present invention are:

Step 1: Design a macroscopic long-term average traffic velocity trajectory prediction module;

It is characterized by:

Step 2: Design a microscopic short-term vehicle speed prediction module

BPNN的结构可以由带权重和阈值的离散模型表示The neural network consists of three layers, namely the input layer, the hidden layer and the output layer. The input vector is defined as m(k) and the output vector is defined as

The structure of BPNN can be represented by a discrete model with weights and thresholds

代表输出的预测车速序列，g(h)是隐含层到输出层的激活函数，其传递函数where w ₁ is the weight between the input layer and the hidden layer, w ₂ is the weight between the hidden layer and the output layer, b ₁ is the threshold of the neurons in the hidden layer, and b ₂ is the threshold of the neurons in the output layer , m(k) represents the input historical speed sequence,

Represents the output predicted speed sequence, g(h) is the activation function from the hidden layer to the output layer, and its transfer function

Step 3: Establish a fuel cell vehicle power system model for energy optimal control

3.1 Establish the vehicle longitudinal driving dynamics model

Fuel cell vehicle parameters: motor transmission efficiency η _{t_veh} (%), mass coefficient of rotating elements σ _veh (-), gravitational acceleration g (m/s ² ), air resistance coefficient C _{D_veh} (-), air density ρ _air (kg /m ³ ), vehicle mass m _veh (kg), windward area A _veh (m ² ), sliding resistance coefficient f(-), road gradient θ _road (-);

demand power of the vehicle

是车辆的速度V_veh对于时间t的微分；where P _{veh_req} is the required power of the vehicle, f is the sliding resistance coefficient, η _{t_veh} is the motor transmission efficiency, σ _veh is the mass coefficient of the rotating element, m _veh is the vehicle mass, g is the acceleration of gravity, θ _road is the road slope, A _veh is the windward area of the car, ρ _air is the air density, C _{D_veh} is the air resistance coefficient,

is the derivative of the vehicle's speed V _veh with respect to time t;

3.2 Establish a fuel cell stack efficiency model

Hydrogen consumption of fuel cells

是氢气的低热值；Among them, P _{fc_req} is the output power of the fuel cell, η _{fc_st} is the working efficiency of the fuel cell,

is the lower calorific value of hydrogen;

3.3 Establish a power battery SOC model

P _{batt_req} =P _{veh_req} -P _{fc_req} (5)

Among them, P _{batt_req} is the output power of the power battery;

The SOC dynamic equation of the power battery is:

是动力电池荷电状态SOC的导数；Among them, V _{oc_batt} is the open circuit voltage of the power battery, R _{int_batt} is the internal resistance of the power battery, Q _batt is the total power of the power battery,

is the derivative of the power battery state of charge SOC;

Step 4: Establish an energy optimal management problem

4.1 The state equation of the output power P _{fc_req} of the fuel cell is:

表示；The above formula can be obtained by the formula

express;

Minimize the hydrogen consumption of the system in the predicted time domain [t ₀ ,t _f ]:

代表系统在t时刻的耗氢量是与t时刻的控制输入u(t)变量有关的函数，控制输入变量取u＝P_{fc_req}，状态变量取x＝SOC，φ(x(t_f))是状态变量的终端约束；Among them, J is the total hydrogen consumption in the prediction time domain under the condition that the system terminal constraints are satisfied, t ₀ is the start time of the prediction time domain, t _f is the end time of the prediction time domain, u is the control input variable, and U is the control input variable. The set of values for the input variable,

It represents that the hydrogen consumption of the system at time t is a function related to the control input u(t) variable at time t. The control input variable takes u=P _{fc_req} , the state variable takes x=SOC, and φ(x(t _f )) is Terminal constraints for state variables;

4.2 The constraints that are satisfied are as follows:

(1) The output power constraints of the fuel cell need to be met:

P _{fc_low ≤P fc_req} (t) _{≤P fc_up} (10)

Among them, P _{fc_low} is the minimum output power of the fuel cell, P _{fc_up} is the maximum output power of the fuel cell, and P _{fc_req} (t) is the output power of the fuel cell at time t;

(2) The dynamic equation and state constraints of the power battery SOC need to be satisfied:

Among them, SOC _begin is the SOC value of the power battery at the initial time, SOC _low is the minimum value of the SOC of the power battery, SOC _up is the maximum value of the SOC of the power battery, SOC(t) represents the value of the SOC of the power battery at time t, SOC(t ₀ ) represents the value of the SOC of the power battery at the initial moment, and SOC(t _f ) represents the value of the SOC of the power battery at the terminal moment;

(3) The power constraints of the power battery need to be met

P _{batt_low ≤P batt_req} (t) _{≤P batt_up} (12)

Among them, P _{batt_low} is the maximum charging power of the power battery, P _{batt_up} is the maximum discharging power of the power battery, and P _{batt_req} (t) is the output power of the power battery at time t;

(4) It is necessary to meet the demand power when the car is running

P _{veh_req} (t)=P _{batt_req} (t)+P _{fc_req} (t) (13)

Among them, P _{veh_req} (t) is the required power of the car at time t, and P _{batt_req} (t) is the output power of the power battery at time t;

Step 5: Using the long-time domain preview information, design the upper-layer SOC trajectory rolling optimization controller

5.1 Upper-layer SOC trajectory rolling optimization controller optimization problem

Discrete the prediction time domain [t _0,m ,t _f,m ] under the time scale into N _m equal parts, where t _0,m is the start time of the prediction time domain, and t _f,m is the prediction time The termination time of the time domain, the discrete time is denoted as k∈{1,2,...,N _m +1}, and the optimization objective is obtained:

代表耗氢量是与k时刻的控制输入u(k)有关的函数，Δt是相邻两车速信息间的采样时间间隔，控制变量u(k)是燃料电池在k时刻的输出功率P_{fc_req_m}(k)；Among them, J is the total hydrogen consumption of the system at all sampling times under the condition of terminal constraints, φ(x(N _m +1)) is the terminal constraints of the state variables,

The representative hydrogen consumption is a function related to the control input u(k) at time k, Δt is the sampling time interval between two adjacent vehicle speed information, and the control variable u(k) is the output power of the fuel cell at time k P _{fc_req_m} ( k);

The specific constraints that are met are:

(1) Satisfy the output power constraints of the fuel cell:

P _{fc_low ≤P fc_req_m} (k) _{≤P fc_up} (15)

(2) Satisfy the dynamic equation and state constraints of the power battery SOC in this time domain:

Among them, SOC _m (k+1) is the value of the SOC of the power battery at the next moment in the time domain obtained after the control input of the power battery SOC _m (k) at time k, that is, the value of the SOC of the power battery at time k+1 , V _{oc_batt_m} (k) is the open circuit voltage of the power battery at time k in this time domain, R _{int_batt_m} (k) is the internal resistance of the power battery at time k in this time domain, P _{veh_req_m} (k) is the required power of the car at time k , P _{fc_req_m} (k) is the output power of the fuel cell at time k in this time domain, SOC _m (k) is the state of charge value of the power battery at time k in this time domain, SOC _m (1) is the power battery in this time domain The value of the battery SOC at the initial time, SOC _m (N _m +1) is the value of the power battery SOC terminal time in this time domain;

(3) Satisfy the output power constraints of the power battery:

P _{batt_low ≤P batt_req_m} (k) _{≤P batt_up} (17)

Among them, P _{batt_req_m} (k) is the output power of the power battery at time k in this time domain;

(4) To meet the demand power when the car is running

P _{veh_req_m} (k)=P _{fc_req_m} (k)+P _{batt_req_m} (k) (18);

5.2 Meshing about system states and control variables

The state variable power battery is divided into 81 state grids; the output power of the fuel cell increases by 81 control variable grids from the beginning;

5.3 Calculate the cost of consideration

计算得到，将每一次从前向后迭代计算产生的代价成本存储在网格中；Under the action of the control variable u(k), the state variable x(k) will be calculated by the state transition equation to obtain a new state variable x(k+1). From time 1, different control variable grids act on the state The state variable grid at the next moment will be obtained on the variable grid, resulting in the corresponding cost J(k), and the new control variable grid will act on the state variable grid at this moment to generate the corresponding cost at the next moment. The cost J(k+1) is calculated until the entire driving cycle is completed, and the resulting cost can be calculated by the formula

After the calculation is obtained, the cost generated by each iteration from the forward to the backward is stored in the grid;

5.4 Determining the optimal decision

Determine the value of the state variable at the terminal moment k=N _m +1, that is, x(N _m +1), corresponding to the initial objective function J(N _m +1)=0, then from the last moment of the terminal moment:

Among them, J ^* (k) represents the minimum value of hydrogen consumption when the system state variable at the kth time is x(k). L(x(k), u(k)) represents the hydrogen consumption generated by the state variable x(k) under the action of the control input u(k) at the k-th moment, that is, the state transition cost, J ^* (k +1) is the minimum value of hydrogen consumption when the system state variable is x(k+1) at the previous moment. Select the state variable corresponding to the minimum cost function from each moment to obtain the optimal state variable Sequence {x ^* (1),x ^* (2),...,x ^* (k)}, namely the optimal power battery SOC sequence SOC ^* ;

Step 6: Use the short-time domain preview information to design the lower-level energy rolling optimization controller

6.1 Receive the SOC ^* trajectory sequence in the predicted time domain at the current sampling time, and read the SOC value of the current power battery;

6.2 The lower energy rolling optimization controller optimization problem

The vehicle speed preview information in the microscopic short-time domain [t _0,n ,t _f,n ] is discretized into N _n equal parts, and the discrete time is denoted as s∈{1,2,...,N _n +1}, where t _0,n is the start time of the prediction time domain, t _f,n is the end time of the prediction time domain, and the optimization objective function is obtained:

代表耗氢量是与控制输入u_d(s)有关的函数，控制变量选取u_d＝P_{fc_req_n}(s)，状态变量选取x_d＝SOC_n(s)，满足的具体约束条件是：Among them, _{ud is the control input in this time domain, U d is} the set of control input values within the range of the control input in this time domain, and I is the SOC of the power battery at all sampling times of the system under the condition that the system terminal constraints are satisfied The sum of the square of the difference with SOC ^* and the sum of hydrogen consumption, SOC _n (s) is the value of the power battery SOC at s time in this time domain, φ(x _d (N _n +1)) is the terminal constraint of the state variable , _ud (s) is the value of the control input at time s in this time domain,

It represents that the hydrogen consumption is a function related to the control input _ud (s), the control variable is selected as _ud =P _{fc_req_n} (s), and the state variable is selected as x _d =SOC _n (s). The specific constraints are:

(1) Satisfy the output power constraints of the fuel cell:

P _{fc_low} <P _{fc_req_n} (s) < P _{fc_up} (21)

(2) Satisfy the dynamic equation and state constraints of the power battery SOC:

Among them, SOC _n (s+1) is the value of the SOC of the power battery at the next moment in the time domain obtained by the SOC _n (s) of the power battery at the time s in the time domain, that is, the power battery at the time s+1. The value of SOC, V _{oc_batt_n} (s) is the open circuit voltage of the power battery at time s in this time domain, P _{veh_req_n} (s) is the required power of the car at time s in this time domain, and P _{fc_req_n} (s) is s in this time domain The output power of the fuel cell at the time, R _{int_batt_n} (s) is the internal resistance of the power battery at the time s in the time domain, SOC _n (1) is the SOC value of the power battery at the initial time in the time domain, SOC _n (N _n +1 ) is the value of the power battery SOC at the terminal moment in this time domain;

(3) Satisfy the output power constraints of the power battery:

P _{batt_low ≤P batt_req_n} (s) _{≤P batt_up} (23)

(4) To meet the demand power when the car is running:

P _{veh_req_n} (s)=P _{fc_req_n} (s)+P _{batt_req_n} (s) (24)

6.3 Constructing the Hamiltonian

H(x _d (s),u _d (s),λ(s),s)=(SOC _n (s)-SOC ^* (s)) ² ·Δt

+((W _{H2_fc} (u _d (s)) Δt+λ(s)ΔSOC _n (s), (25)

Among them, H(x _d (s), u _d (s), λ(s), s) represents the value x _d (s) of the Hamiltonian function and state variables at time s, and the value of control input at time s _{d d} (s), the value λ(s) of the co-state variable at time s is related to the current time s, and the necessary conditions to be satisfied for optimization are as follows:

λ(s+1)=λ(s)+Δλ(s)·Δt, (26)

代表s时刻哈密顿函数对动力电池SOC求偏导的值，λ(s+1)是在s时刻协态变量λ(s)经计算后得到的该时域下下一时刻协态变量的值，即协态变量在s+1时刻的值。同时，最优控制输入

需要在每个采样时刻保证哈密顿函数最小，即需要满足以下公式：Among them, Δλ(s) is the difference between the co-state variables at two adjacent moments,

Represents the value of the partial derivative of the Hamiltonian function at time s to the SOC of the power battery, λ(s+1) is the value of the co-state variable at the next time in the time domain obtained by calculating the co-state variable λ(s) at time s , that is, the value of the covariate at time s+1. At the same time, the optimal control input

It is necessary to ensure the minimum Hamiltonian function at each sampling moment, that is, the following formula needs to be satisfied:

是状态变量在s时刻的最优值，

是控制输入在s时刻的最优值，λ^*(s)是协态变量在s时刻的最优值，

代表最优哈密顿函数与状态变量在s时刻的最优值

控制输入在s时刻的最优值

协态变量在s时刻的最优值λ^*(s)和当前时刻s有关，

代表哈密顿函数与状态变量在s时刻的最优值

控制输入在s时刻的值u_d(s)、协态变量在s时刻的最优值λ^*(s)和当前时刻s有关；in,

is the optimal value of the state variable at time s,

is the optimal value of the control input at time s, λ ^* (s) is the optimal value of the covariate variable at time s,

Represents the optimal Hamiltonian function and the optimal value of the state variable at time s

The optimal value of the control input at time s

The optimal value λ ^* (s) of the covariate at time s is related to the current time s,

Represents the optimal value of the Hamiltonian function and the state variable at time s

The value _ud (s) of the control input at time s, and the optimal value λ ^* (s) of the co-state variable at time s are related to the current time s;

6.4 Solving the Optimal Control Input Sequence

(1) Set the state variable SOC ₀ at the initial moment, and calculate the initial value λ ₀ of the co-state variable at the initial moment by the bisection method;

(2) The control input when the Hamiltonian function value is the smallest is the optimal control input

U _d (s)=[u _{d_low} (s):Δu _d (s):u _{d_up} (s)], (28)

Among them, _Δud (s) is the difference between two adjacent equal parts of the control input at time s in the time domain, _{ud_up} (s) is the maximum value of the control input within the constraint range at time s, and _{ud_low} (s) is The minimum value of the control input within the constraint range, U _d (s) is the set of values of the control input at time s in this time domain;

在状态转移方程的结果，计算下一采样时刻的状态变量SOC值及协态变量λ的值，重复步骤2)直至最后一个采样时刻；(3) According to the optimal control input action

At the result of the state transition equation, calculate the value of the state variable SOC and the value of the co-state variable λ at the next sampling time, and repeat step 2) until the last sampling time;

(4) Judging the error value of the last terminal boundary of the SOC value, if the error is within the set range, end the calculation, otherwise it is necessary to re-input λ ₀ , and within the value range set by λ to determine the The value of the co-state variable λ within the allowable error range, repeat step 2), and the optimal control input sequence can be obtained after all calculations are completed;

Step 7: Transmit the obtained control input sequence signal to the power execution control unit of the fuel cell vehicle.

The invention is oriented towards the fuel cell vehicle in the intelligent network connection environment, and proposes a real-time energy optimization management system for the intelligent network connection fuel cell vehicle, which solves the contradiction between the optimization real-time performance and the energy saving effect existing in the existing energy optimization research algorithm, and further explores the intelligent network. The energy-saving space of fuel cell vehicles in the networked transportation environment significantly improves the fuel economy of fuel cell vehicles in the intelligent networked environment.

Description of drawings

Figure 1 is a structural diagram of the power and transmission part of a fuel cell vehicle;

Figure 2 is a schematic diagram of the real-time energy optimization management system for intelligent networked fuel cell vehicles;

Fig. 3 is the working flow chart of the real-time energy optimization management system of the intelligent networked fuel cell vehicle;

Figure 4 is a schematic diagram of the BPNN algorithm predicting short-term vehicle speed;

FIG. 5 is a schematic diagram of multi-time scale vehicle speed information prediction;

Fig. 6 is the working efficiency curve diagram of the fuel cell;

Figure 7 is a graph showing the relationship between the internal resistance of the power battery and SOC;

Fig. 8 is the vehicle speed curve diagram of the selected congestion condition driving cycle (LA92);

Figure 9 shows the optimal SOC ^* trajectory of the power battery calculated by the upper-layer SOC trajectory rolling optimization controller under the congested driving cycle (LA92);

Figure 10 is a comparison diagram of the vehicle speed curve of the congested driving cycle (LA92) and the vehicle speed curve predicted by BPNN;

Figure 11 is a comparison diagram of the reference optimal SOC ^* trajectory calculated by the upper-layer SOC trajectory rolling optimization controller under the congested driving cycle (LA92) and the actual SOC curve calculated by the lower-layer energy rolling optimization controller;

Figure 12 is a graph showing the fuel cell output power and the power battery output power calculated by the vehicle's required power and the underlying energy rolling optimization controller under the congested driving cycle (LA92);

Fig. 13 is the calculation time curve required to solve the output power of the fuel cell and the power battery at each sampling time of the energy rolling optimization controller under the congested driving cycle (LA92);

Figure 14 is the hydrogen consumption curve calculated by the real-time energy optimization management system of the intelligent networked fuel cell vehicle under the driving cycle (LA92) under the congested condition;

Figure 15 shows the hydrogen consumption calculated by the real-time energy optimization management system of the intelligent networked fuel cell vehicle under the congested driving cycle (LA92), the hydrogen consumption calculated by the rule-based energy management strategy, and the offline global optimal hydrogen consumption Result comparison chart.

Detailed ways

The main challenges and problems of mining fuel cell vehicles in the intelligent networked transportation environment include: 1. The macro and micro traffic preview information obtained in the networked environment is dynamically updated with the migration of time and space, how to use the rich and multi-source information It is a challenge to design an efficient energy optimization management system to realize the predicted energy saving of fuel cell vehicles; 2. Frequent starting and stopping of fuel cells will lead to performance degradation, and the state of charge (SOC) of the power battery is only in the High working efficiency can only be achieved within a suitable working range. During the driving process of a fuel cell hybrid vehicle, the start and stop of the fuel cell stack and the range of the SOC of the power battery need to be strictly limited. Design a real-time multi-power source with constraints The energy optimization algorithm is another challenge.

The steps of the present invention are:

Step 1: Design a macroscopic long-term average traffic velocity trajectory prediction module. The network terminal collects the driving information of electronic maps, Global Positioning System (GPS) and other vehicles in networked traffic, and uses cloud computing resources to generate macroscopic long-term (600 seconds) average traffic velocity trajectory preview information , and dynamically updated in real time according to the frequency of once every 300 seconds.

Step 2: Design a microscopic short-term vehicle speed prediction module. Combined with the average traffic velocity trajectory information in the macro-long-term domain, and based on the historical vehicle speed sequence of the vehicle in the past 5 seconds, the Back Propagation Neural Network (BPNN) algorithm is used to generate a preview of the micro-short-time domain (5 seconds). The speed information is dynamically updated in real time according to the frequency of once a second.

Step 3: Establish a fuel cell vehicle power system model for energy optimal control. The vehicle longitudinal driving dynamics model, the fuel cell stack efficiency model, and the power battery SOC model are established; the required power of the vehicle is calculated according to the macroscopic long-term average traffic velocity trajectory information and the microscopic short-term vehicle speed preview information.

Step 4: Establish a description of the energy optimal management problem. The control input variables are selected, the energy optimization management problem description is established, and the constraints of the optimization problem are determined.

Step 5: Use the long-time domain preview information to design the upper-layer SOC trajectory rolling optimization controller. According to the macroscopic long-term average traffic velocity trajectory information provided by cloud computing resources and the required power calculated in step 2, and under the condition that the system terminal constraints are met, the goal is to minimize the hydrogen consumption of fuel cell vehicles. The online solution algorithm of the rolling optimization problem of the planning algorithm (Dynamic Programming, DP) uses cloud computing resources to solve the optimal power battery SOC trajectory in this time domain.

Step 6: Use the short-time domain preview information to design the lower-level energy rolling optimization controller. The power battery reference SOC trajectory (SOC ^* ) optimized by the SOC trajectory rolling optimization controller is used as the reference input of the lower energy rolling optimization controller. While ensuring that the system terminal constraints are met, the actual power battery SOC in the microscopic short time domain is used to compare The SOC ^* trajectory tracking error and the minimum hydrogen consumption given by the upper-layer SOC trajectory rolling optimization controller are the goals. Under the theoretical framework of the Pontryagin Maximum Principle (PMP), a fuel cell vehicle is proposed. The online solution algorithm for the energy rolling optimization problem solves the output power sequence of the fuel cell and the power battery in the short time domain, that is, the optimal control input sequence of the system.

Step 7: Transmit the obtained control input sequence signal to the power execution control unit of the fuel cell vehicle.

Step 8: Carry out experimental simulation to evaluate the energy saving effect of the designed real-time energy optimization management system.

The information collected by the present invention includes location information and slope information collected through GPS, route information collected through electronic maps, intersection traffic light information collected through network communication, and driving status information of other vehicles.

The invention is oriented to the fuel cell vehicle in the intelligent networked environment, uses the traffic preview information provided by cloud computing resources, and considers the power characteristics of the fuel cell and the problem of performance degradation caused by frequent starting and stopping, and proposes the real-time energy of the intelligent networked fuel cell vehicle. Optimize the management system and tap the hydrogen-saving potential of fuel cell vehicles in the intelligent networked transportation environment.

The multi-scale dynamic traffic preview network connection information involved in the present invention includes: macroscopic long-time domain (minute level) average traffic velocity trajectory information and microscopic short-time domain (second level) vehicle speed preview information. Among them, the average traffic velocity trajectory information in the long-term domain is processed by the cloud computing data processing center, provided to the fuel cell vehicle, and dynamically updated at a fixed time (usually 5-10 minutes). The vehicle speed trajectory preview information in the short time domain (usually 5-10 seconds) can be combined with the historical vehicle speed sequence of the target vehicle and the average traffic velocity trajectory information in the long time domain and other data through the machine learning method (the present invention takes BPNN as an example) in real time Learn to get.

The invention combines the traffic information of macro long time domain and micro short time domain with two different time scales, aiming at the energy coordination optimization problem of multiple power sources of fuel cell vehicles, and considering the system safety constraints in the congested driving state, a layered fuel cell vehicle is proposed. A real-time energy rolling optimization control method, which includes 1) an upper-layer SOC trajectory rolling optimization controller and 2) a lower-layer energy rolling optimization controller, described as follows:

(1) Upper-layer SOC trajectory rolling optimization controller: Using cloud computing resources, the average traffic velocity trajectory information in the front 600 seconds, which is updated every 300 seconds, is used as the reference input of the system. The optimum is the goal, and the DP algorithm is applied to solve the SOC trajectory SOC ^* of the power battery under the condition of minimum hydrogen consumption.

(2) Lower-layer energy rolling optimization controller: The vehicle-mounted controller combines the upper-layer average traffic velocity trajectory information, the traffic light information at the intersection, the slope information of the current road and the historical vehicle speed sequence of the past 5 seconds, and uses the BPNN algorithm to obtain every 1 second The frequency of one time predicts the microscopic short-term vehicle speed sequence 5 seconds into the future. Combined with the obtained short-term vehicle speed prediction sequence, the on-board controller takes the optimal power battery SOC trajectory SOC ^* obtained by the upper-layer SOC trajectory rolling optimization controller as the reference input of this layer. The power battery SOC to the upper-layer SOC trajectory rolling optimization controller gives the SOC ^* trajectory tracking error and the minimum hydrogen consumption as the goal. In this short time domain, the PMP algorithm is used to solve the output power of the fuel cell and the power battery, and the optimal power distribution scheme.

The structure diagram of the power and transmission part of the fuel cell vehicle of the present invention is shown in Figure 1. The fuel cell and the power battery are respectively connected to the circuit bus through a one-way DC/DC converter and a two-way DC/DC converter, and the circuit bus and the motor are connected between Connected through a bidirectional DC/AC converter, the motor drives the wheels to rotate, providing power for the car to drive.

The schematic diagram of the real-time energy optimization management system of the intelligent network-connected fuel cell vehicle of the present invention is shown in FIG. 2 . The specific implementation is as follows: the network terminal collects information such as the location information of the car, the destination information and the interaction information of other connected vehicles and uploads it to the cloud computing data processing center. After the cloud computing data processing center processes the collected information, it generates macro-long-term average traffic velocity trajectory information (600 seconds) according to the update frequency of once every 300 seconds and sends it to the target vehicle. The on-board controller combines the average traffic velocity of the upper layer. Trajectory information, traffic light information at intersections, current road gradient information, and historical vehicle speed sequences in the past 5 seconds, the BPNN algorithm is used to predict the micro-short-term vehicle speed sequence in the future 5 seconds according to the frequency of updating every 1 second. Then the vehicle longitudinal driving dynamics model, the fuel cell stack efficiency model, the battery SOC model and the real-time energy optimization management system are established, and the problem description of the real-time energy optimization management problem is established to determine the constraints of the optimization problem. In the real-time energy optimization management system, the upper-layer SOC trajectory rolling optimization controller combines the average traffic velocity information and demand power in the macro-long time domain of 600 seconds ahead, which is updated every 300 seconds, to calculate the hydrogen consumption under the condition that the system terminal constraints are met. The minimum is the goal, and the cloud computing resource uses the DP algorithm to solve the SOC trajectory SOC ^* of the power battery in this time domain. The lower-layer energy rolling optimization controller takes the optimal SOC trajectory (SOC ^* ) as the reference input, and the on-board controller guarantees to satisfy the system terminal constraints while using the power battery SOC in the microscopic short-time domain to the upper-layer SOC trajectory rolling optimization controller. The SOC ^* trajectory tracking error and the minimum hydrogen consumption are the goals, and the fuel cell power and power battery power sequence of the system are obtained, that is, the optimal control input sequence. Finally, the obtained optimal control input sequence signal is transmitted to the power execution control unit of the fuel cell vehicle. The experimental simulation of the designed system is carried out to verify the energy-saving effect of the designed system on fuel cell vehicles. specifically:

The working flow chart of the real-time energy optimization management system of the intelligent networked fuel cell vehicle is shown in Figure 3, which includes the following steps:

1. Design a macro-long-term average traffic velocity trajectory prediction module.

The network terminal collects the driving information of the electronic map, GPS and other vehicles in the networked traffic, and uses the cloud computing resources to generate the predicted macroscopic long-term (600 seconds) average traffic velocity trajectory information, and transmit it to the fuel cell vehicle. It is dynamically updated in real time at a frequency of once every 300 seconds.

2. Design a vehicle speed prediction module in the microscopic short-term domain

The on-board controller combines the macroscopic long-term average traffic velocity trajectory information, the traffic light information at the intersection, the current road gradient information and the historical vehicle speed sequence of the past 5 seconds, and uses the BPNN algorithm to generate the microscopic short-term (5 seconds) vehicle speed. Preview information and update it dynamically in real time at a frequency of once a second.

BPNN的结构可以由带权重和阈值的离散模型表示，如公式(1)所示：The schematic diagram of the BPNN algorithm for predicting short-term vehicle speed is shown in Figure 4. The neural network consists of three layers, namely the input layer, the hidden layer and the output layer. The input vector is defined as m(k) and the output vector is defined as

The structure of BPNN can be represented by a discrete model with weights and thresholds, as shown in formula (1):

代表输出的预测车速序列，g(h)是隐含层到输出层的激活函数。where w ₁ is the weight between the input layer and the hidden layer, w ₂ is the weight between the hidden layer and the output layer, b ₁ is the threshold of the neurons in the hidden layer, and b ₂ is the threshold of the neurons in the output layer , m(k) represents the input historical speed sequence,

represents the output predicted vehicle speed sequence, and g(h) is the activation function from the hidden layer to the output layer.

Its transfer function is shown in formula (2):

The neural network vehicle speed prediction model is obtained after training on a variety of typical operating conditions. In this model, the historical vehicle speed sequence of 5 seconds is input at the frequency of once every 1 second, and the vehicle speed sequence of the next 5 seconds can be predicted in a rolling manner. Figure 10 is a comparison chart of the vehicle speed curve of the driving cycle (LA92) under the congestion condition and the vehicle speed curve predicted by BPNN. It can be seen from the figure that the vehicle speed curve predicted by the BPNN algorithm can be highly consistent with the actual driving cycle speed curve, and has good prediction effect.

3. Establish a fuel cell vehicle power system model for energy optimization control.

Establish vehicle longitudinal driving dynamics model, fuel cell stack efficiency model, power

The pool SOC model combines the average traffic velocity trajectory information in the macroscopic long-term domain and the vehicle speed preview information in the microscopic short-term domain to calculate the required power of the vehicle.

3.1 Establish the vehicle longitudinal driving dynamics model

According to the vehicle speed preview information, the required power when the vehicle is running is calculated. The fuel cell vehicle parameters of this patent are shown in Table 1:

Table 1: Fuel cell vehicle parameter table

Variable meaning symbols and units Motor drive efficiency η_{t_veh}(%) Mass factor of rotating element σ_veh(-) Gravitational acceleration g(m/s²) Air drag coefficient C_{D_veh}(-) Air density ρ_air(kg/m³) car quality m_veh(kg) Frontal area A_veh(m²) Sliding resistance coefficient f(-) road slope θ_road(-)

According to the macroscopic long-term average traffic velocity trajectory information provided by the cloud computing resources, the required power of the vehicle is calculated by formula (3).

是车辆的速度V_veh对于时间t的微分。where P _{veh_req} is the required power of the vehicle, f is the sliding resistance coefficient, η _{t_veh} is the motor transmission efficiency, σ _veh is the mass coefficient of the rotating element, m _veh is the vehicle mass, g is the acceleration of gravity, θ _road is the road slope, A _veh is the windward area of the car, ρ _air is the air density, C _{D_veh} is the air resistance coefficient,

is the derivative of the vehicle's speed V _veh with respect to time t.

3.2 Establish a fuel cell stack efficiency model

计算公式如下：The fuel cell is the main energy source for driving the fuel cell vehicle, and the working efficiency curve of the fuel cell involved in the present invention is obtained by means of experimental calibration, as shown in FIG. 6 . Hydrogen consumption of fuel cells

Calculated as follows:

是氢气的低热值，数值为120000J/g，它表示燃烧1g氢气会产生120000J的能量。Among them, P _{fc_req} is the output power of the fuel cell, η _{fc_st} is the working efficiency of the fuel cell,

Is the low calorific value of hydrogen, the value is 120000J/g, it means that burning 1g of hydrogen will produce 120000J of energy.

3.3 Establish a power battery SOC model

The power battery is the auxiliary energy source of the fuel cell vehicle, and its output power is the difference between the power demanded by the vehicle and the output power of the fuel cell, as shown in formula 5:

P _{batt_req} =P _{veh_req} -P _{fc_req} , (5)

Among them, P _{batt_req} is the output power of the power battery.

The SOC dynamic equation of the power battery is:

是动力电池荷电状态SOC的导数。Among them, V _{oc_batt} is the open circuit voltage of the power battery, R _{int_batt} is the internal resistance of the power battery, Q _batt is the total power of the power battery,

is the derivative of the power battery state of charge SOC.

4. Establish energy optimization management problem description

The control input variables are selected, the energy optimization management problem description is established, and the constraints of the optimization problem are determined.

4.1 Establishment of energy optimization management problem description

The relationship between the internal resistance and SOC of the power battery is shown in Figure 7. When the SOC state is too high or too low, the internal resistance is large and the working efficiency is low; Start-stop is attenuated. Combined with the characteristics of the above-mentioned power sources, it is necessary to strictly limit the start and stop of the fuel cell stack and the range of the SOC of the power battery during the optimization process. In the present invention, the SOC of the power battery is selected as the state variable, and through the analysis of the working principle of the fuel cell hybrid power system, the open circuit voltage V _{oc_batt} and the internal resistance R _{int_batt} of the power battery are both functions related to the SOC of the power battery. Multiple control variables in the dynamic equation are simplified to one control variable, that is, the output power P _{fc_req} of the fuel cell, and its state equation is:

表示。The above formula can be obtained by the formula

express.

The optimization objective is to minimize the hydrogen consumption of the system in the predicted time domain [t ₀ ,t _f ] under the condition of satisfying the system terminal constraints:

代表系统在t时刻的耗氢量是与t时刻的控制输入u(t)变量有关的函数，控制输入变量取u＝P_{fc_req}，状态变量取x＝SOC，φ(x(t_f))是状态变量的终端约束。Among them, J is the total hydrogen consumption in the prediction time domain under the condition that the system terminal constraints are satisfied, t ₀ is the start time of the prediction time domain, t _f is the end time of the prediction time domain, u is the control input variable, and U is the control input variable. The set of values for the input variable,

It represents that the hydrogen consumption of the system at time t is a function related to the control input u(t) variable at time t. The control input variable takes u=P _{fc_req} , the state variable takes x=SOC, and φ(x(t _f )) is Terminal constraints for state variables.

4.2 Determine the constraints of the optimization problem

The constraints that the real-time energy optimization management system of the intelligent networked fuel cell vehicle needs to meet are as follows:

(1) The output power constraints of the fuel cell need to be met:

P _{fc_low ≤P fc_req} (t) _{≤P fc_up} , (9)

Among them, P _{fc_low} is the minimum output power of the fuel cell, P _{fc_up} is the maximum output power of the fuel cell, and P _{fc_req} (t) is the output power of the fuel cell at time t.

The dynamic equation and state constraints of the power battery SOC need to be satisfied:

Among them, SOC _begin is the SOC value of the power battery at the initial time, SOC _low is the minimum value of the SOC of the power battery, SOC _up is the maximum value of the SOC of the power battery, SOC(t) represents the value of the SOC of the power battery at time t, SOC(t ₀ ) represents the value of the SOC of the power battery at the initial moment, and SOC(t _f ) represents the value of the SOC of the power battery at the terminal moment;

4.3 Need to meet the power constraints of the power battery

P _{batt_low ≤P batt_req} (t) _{≤P batt_up} , (11)

Among them, P _{batt_low} is the maximum charging power of the power battery, P _{batt_up} is the maximum discharging power of the power battery, and P _{batt_req} (t) is the output power of the power battery at time t;

4.4 Need to meet the demand power when the car is running

P _{veh_req} (t)=P _{batt_req} (t)+P _{fc_req} (t), (12)

Among them, P _{veh_req} (t) is the required power of the car at time t, and P _{batt_req} (t) is the output power of the power battery at time t.

Combined with the optimization problems and constraints determined above, the optimal control variables of the system are solved. The schematic diagram of multi-time-scale vehicle speed information prediction is shown in Figure 5. In the upper layer, cloud computing resources predict the average traffic velocity trajectory information in the macro-long time domain for 600 seconds ahead at the frequency of once every 300 seconds, and transmit the predicted traffic velocity trajectory information to the upper-layer SOC trajectory rolling optimization controller and the lower-layer on-board vehicle The controller; in the lower layer, the vehicle-mounted controller combines the traffic velocity trajectory information given by the upper layer, the traffic light information at the intersection, the slope information of the current road and the historical vehicle speed sequence of the past 5 seconds, and predicts the front 5 according to the frequency of once every 1 second. Second microscopic short-term vehicle speed preview information, and transfer the short-term vehicle speed preview information to the lower-level energy rolling optimization controller. The upper layer SOC trajectory rolling optimization controller (step 5) and the lower layer energy rolling optimization controller (step 6) are designed according to the vehicle speed preview information of the upper and lower layers with different time scales. The details are as follows:

5. Using the long-time domain preview information, the upper-layer SOC trajectory rolling optimization controller is designed.

Combined with the average traffic velocity trajectory information and required power in the macroscopic long-term domain, aiming at the optimal fuel economy, an upper-layer SOC trajectory rolling optimization controller is designed to solve the optimal SOC trajectory SOC ^* of the power battery in this time domain. Specifically includes the following steps:

5.1 Description of the upper-layer SOC trajectory rolling optimization controller optimization problem

The upper-layer SOC trajectory rolling optimization controller uses the cloud computing resources to update the average traffic velocity trajectory information in the macroscopic long-term domain for 600 seconds ahead, which is updated every 300 seconds. Use cloud computing resources to optimize solutions. In the solution process, it is necessary to sample the data, and discretize the prediction time domain [t _0,m ,t _f,m ] under the time scale into N _m equal parts, where t _0,m is the prediction time domain of the prediction time domain [t 0,m ,t f,m ] The starting time, t _f,m is the end time of the prediction time domain, and the discrete time is denoted as k∈{1,2,...,N _m +1}, and the optimization objective is obtained:

代表耗氢量是与k时刻的控制输入u(k)有关的函数，Δt是相邻两车速信息间的采样时间间隔，控制变量u(k)是燃料电池在k时刻的输出功率P_{fc_req_m}(k)，轨迹滚动优化过程中需要保证动力电池初始的SOC值等于终端的SOC值，因为当动力电池初始的SOC与最终的SOC不同时，会产生等效氢消耗。如最终的SOC值比初始SOC值高时，认为燃料电池的一部分能量存储在动力电池中，并没有真的被消耗，同理，当动力电池最终的SOC值比初始的SOC值低时，认为动力电池替燃料电池额外消耗了能量。为避免动力电池产生额外的等效氢消耗的对耗氢量结果造成的干扰，便于对比耗氢量结果，需要约束最终的SOC与初始的SOC值相同。需要满足的具体约束条件是：Among them, J is the total hydrogen consumption of the system at all sampling times under the condition of terminal constraints, φ(x(N _m +1)) is the terminal constraints of the state variables,

The representative hydrogen consumption is a function related to the control input u(k) at time k, Δt is the sampling time interval between two adjacent vehicle speed information, and the control variable u(k) is the output power of the fuel cell at time k P _{fc_req_m} ( k), in the process of trajectory rolling optimization, it is necessary to ensure that the initial SOC value of the power battery is equal to the terminal SOC value, because when the initial SOC of the power battery is different from the final SOC, equivalent hydrogen consumption will occur. If the final SOC value is higher than the initial SOC value, it is considered that a part of the energy of the fuel cell is stored in the power battery and is not really consumed. Similarly, when the final SOC value of the power battery is lower than the initial SOC value, it is considered that The power battery consumes additional energy for the fuel cell. In order to avoid the interference on the hydrogen consumption results caused by the additional equivalent hydrogen consumption of the power battery, and to facilitate the comparison of the hydrogen consumption results, it is necessary to constrain the final SOC to be the same as the initial SOC value. The specific constraints that need to be met are:

1. The output power constraints of the fuel cell need to be met:

P _{fc_low ≤P fc_req_m} (k) _{≤P fc_up} ; (14)

2. The dynamic equation and state constraints of the power battery SOC in this time domain need to be satisfied:

Among them, SOC _m (k+1) is the value of the SOC of the power battery at the next moment in the time domain obtained after the control input of the power battery SOC _m (k) at time k, that is, the value of the SOC of the power battery at time k+1 , V _{oc_batt_m} (k) is the open circuit voltage of the power battery at time k in this time domain, R _{int_batt_m} (k) is the internal resistance of the power battery at time k in this time domain, P _{veh_req_m} (k) is the required power of the car at time k , P _{fc_req_m} (k) is the output power of the fuel cell at time k in this time domain, SOC _m (k) is the state of charge value of the power battery at time k in this time domain, SOC _m (1) is the power battery in this time domain The value of the battery SOC at the initial time, SOC _m (N _m +1) is the value of the power battery SOC terminal time in this time domain;

3. The output power constraints of the power battery need to be met:

P _{batt_low ≤P batt_req_m} (k) _{≤P batt_up} , (16)

Among them, P _{batt_req_m} (k) is the output power of the power battery at time k in this time domain;

4. Need to meet the demand power when the car is running

P _{veh_req_m} (k)=P _{fc_req_m} (k)+P _{batt_req_m} (k), (17)

5.2 Meshing about system states and control variables

In order to apply the DP method, the SOC of the state variable power battery is increased from 0.3 to 0.7 in increments of 0.005 per grid, and 81 state grids are divided; the output power of the fuel cell is increased from 5kW to 50kW in increments of 0.5625kW per grid A grid of 81 control variables.

5.3 Calculate the cost of consideration

计算得到，将每一次从前向后迭代计算产生的代价成本存储在网格中。Under the action of the control variable u(k), the state variable x(k) will be calculated by the state transition equation to obtain a new state variable x(k+1). Starting from time 1, different control variable grids act on the state variable grid to obtain the state variable grid at the next moment, resulting in the corresponding cost J(k). At the same time, the new control variable grid acts on the state variable grid at this moment, and the cost J(k+1) corresponding to the next moment is generated until the calculation of the entire driving cycle is completed. The resulting cost can be calculated by the formula

The calculation is obtained, and the cost of each iteration from forward to backward is stored in the grid.

5.4 Determining the optimal decision

Determining the optimal control input sequence needs to be realized by iterative calculation from back to front. The optimization objective of the optimization problem is to minimize the hydrogen consumption of the fuel cell vehicle under the condition of satisfying the terminal constraints. First determine the value of the state variable at the terminal time k=N _m +1 (the value is SOC _begin ), that is, x(N _m +1), corresponding to the initial objective function J(N _m +1)=0, then from the terminal time The last moment starts with:

Among them, J ^* (k) represents the minimum value of hydrogen consumption when the system state variable at the kth time is x(k). L(x(k), u(k)) represents the hydrogen consumption generated by the state variable x(k) under the action of the control input u(k) at the k-th moment, that is, the state transition cost. J ^* (k+1) is the minimum hydrogen consumption when the system state variable is x(k+1) at the previous moment. Selecting the state variable corresponding to the minimum cost function at each moment, the optimal state variable sequence {x ^* (1),x ^* (2),...,x ^* (k)} can be obtained, That is, the optimal power battery SOC sequence SOC ^* .

6. Use the short-time domain preview information to design the lower-level energy rolling optimization controller.

The optimal SOC trajectory SOC ^* given by the upper-layer SOC trajectory rolling optimization controller is used as the reference input, combined with the vehicle speed preview information in the micro-short-time domain of 5 seconds ahead, which is updated every 1 second and obtained by the BPNN algorithm, in order to ensure that the system is satisfied. In addition to the terminal constraints, the actual power battery SOC _n in the microscopic short time domain is aimed at the SOC ^* trajectory tracking error and the minimum hydrogen consumption given by the upper-layer SOC trajectory rolling optimization controller, and the lower-layer energy rolling optimization controller is designed. The controller solves the output power sequence of the fuel cell and power battery in this time domain, that is, the optimal control input sequence of the system. Specifically includes the following steps:

6.1 Receive the SOC ^* trajectory sequence in the predicted time domain at the current sampling time, and read the SOC value of the current power battery.

6.2 Description of the optimization problem of the lower energy rolling optimization controller

The lower-layer energy rolling optimization controller, while ensuring that the system terminal constraints are met, takes the actual power battery SOC _n in the microscopic short-time domain to the SOC ^* trajectory tracking error and the minimum hydrogen consumption given by the upper-layer SOC trajectory rolling optimization controller. , to solve the output power of the fuel cell and power battery. In the solution process, the vehicle speed preview information in the microscopic short-time domain [t _0,n ,t _f,n ] needs to be discretized into N _n equal parts, and the discrete time is denoted as s∈{1,2,...,N _n +1}, where t _0,n is the start time of the prediction time domain, t _f,n is the end time of the prediction time domain, and the optimization objective function is obtained:

代表耗氢量是与控制输入u_d(s)有关的函数，控制变量选取u_d＝P_{fc_req_n}(s)，状态变量选取x_d＝SOC_n(s)。Among them, _{ud is the control input in this time domain, U d is} the set of control input values within the range of the control input in this time domain, and I is the SOC of the power battery at all sampling times of the system under the condition that the system terminal constraints are satisfied The sum of the square of the difference with SOC ^* and the sum of hydrogen consumption, SOC _n (s) is the value of the power battery SOC at s time in this time domain, φ(x _d (N _n +1)) is the terminal constraint of the state variable , _ud (s) is the value of the control input at time s in this time domain,

The representative hydrogen consumption is a function related to the control input _ud (s), the control variable is selected as _ud =P _{fc_req_n} (s), and the state variable is selected as x _d =SOC _n (s).

The specific constraints that need to be met are:

(1) The output power constraints of the fuel cell need to be met:

P _{fc_low} <P _{fc_req_n} (s) < P _{fc_up} ; (20)

(2) The dynamic equation and state constraints of the power battery SOC need to be satisfied:

Among them, SOC _n (s+1) is the value of the SOC of the power battery at the next moment in the time domain obtained by the SOC _n (s) of the power battery at the time s in the time domain, that is, the power battery at the time s+1. The value of SOC, V _{oc_batt_n} (s) is the open circuit voltage of the power battery at time s in this time domain, P _{veh_req_n} (s) is the required power of the car at time s in this time domain, and P _{fc_req_n} (s) is s in this time domain The output power of the fuel cell at the time, R _{int_batt_n} (s) is the internal resistance of the power battery at the time s in the time domain, SOC _n (1) is the SOC value of the power battery at the initial time in the time domain, SOC _n (N _n +1 ) is the SOC value of the power battery at the terminal time in this time domain, and the energy rolling optimization controller must also satisfy that the SOC value of the system terminal is equal to the initial SOC value to avoid the interference caused by equivalent hydrogen consumption;

3. The output power constraints of the power battery need to be met:

P _{batt_low ≤P batt_req_n} (s) _{≤P batt_up} ; (22)

4. Need to meet the demand power when the car is running:

P _{veh_req_n} (s)=P _{fc_req_n} (s)+P _{batt_req_n} (s). (23)

6.3 Constructing the Hamiltonian

The Hamiltonian function is constructed as follows:

Among them, H(x _d (s), u _d (s), λ(s), s) represents the value x _d (s) of the Hamiltonian function and state variables at time s, and the value of control input at time s _{d d} (s), the value λ(s) of the covariate at time s is related to the current time s.

The necessary conditions to be satisfied for optimization are as follows:

λ(s+1)=λ(s)+Δλ(s)·Δt,

代表s时刻哈密顿函数对动力电池SOC求偏导的值，λ(s+1)是在s时刻协态变量λ(s)经计算后得到的该时域下下一时刻协态变量的值，即协态变量在s+1时刻的值。Among them, Δλ(s) is the difference between the co-state variables at two adjacent moments,

Represents the value of the partial derivative of the Hamiltonian function at time s to the SOC of the power battery, λ(s+1) is the value of the co-state variable at the next time in the time domain obtained by calculating the co-state variable λ(s) at time s , that is, the value of the covariate at time s+1.

需要在每个采样时刻保证哈密顿函数最小，即需要满足以下公式：At the same time, the optimal control input

It is necessary to ensure the minimum Hamiltonian function at each sampling moment, that is, the following formula needs to be satisfied:

是状态变量在s时刻的最优值，

是控制输入在s时刻的最优值，λ^*(s)是协态变量在s时刻的最优值，

代表最优哈密顿函数与状态变量在s时刻的最优值

控制输入在s时刻的最优值

协态变量在s时刻的最优值λ^*(s)和当前时刻s有关，

代表哈密顿函数与状态变量在s时刻的最优值

控制输入在s时刻的值u_d(s)、协态变量在s时刻的最优值λ^*(s)和当前时刻s有关，公式(24)～(26)实际上是一个两固定端点的边界值问题，该问题通过满足PMP的必要条件进行求解，得到最优的控制输入序列。in,

is the optimal value of the state variable at time s,

is the optimal value of the control input at time s, λ ^* (s) is the optimal value of the covariate variable at time s,

Represents the optimal Hamiltonian function and the optimal value of the state variable at time s

The optimal value of the control input at time s

The optimal value λ ^* (s) of the covariate at time s is related to the current time s,

Represents the optimal value of the Hamiltonian function and the state variable at time s

The value _ud (s) of the control input at time s and the optimal value λ ^* (s) of the co-state variable at time s are related to the current time s. Formulas (24) to (26) are actually two fixed endpoints. Boundary value problem, the problem is solved by satisfying the necessary conditions of PMP, and the optimal control input sequence is obtained.

6.4 Solving the Optimal Control Input Sequence

(1) Set the state variable SOC ₀ at the initial moment, and calculate the initial value λ ₀ of the co-state variable at the initial moment by the bisection method.

(2) Within the allowable range of the control input, divide the control input into 100 equal parts, and calculate the Hamiltonian function at each sampling time. The control input with the smallest value of the Hamiltonian function is the optimal control input.

U _d (s)=[u _{d_low} (s):Δu _d (s):u _{d_up} (s)],

Among them, _Δud (s) is the difference between two adjacent equal parts of the control input at time s in the time domain, _{ud_up} (s) is the maximum value of the control input within the constraint range at time s, and _{ud_low} (s) is The minimum value of the control input within the constraint range, U _d (s) is the set of values of the control input at time s in this time domain.

在状态转移方程的结果，计算下(3) According to the optimal control input action

As a result of the state transition equation, calculate under

The value of the state variable SOC and the value of the co-state variable λ at a sampling moment, repeat step 2) until the last sampling moment;

(4) Judging the error value of the last terminal boundary of the SOC value, if the error is within the set range, end the calculation, otherwise it is necessary to re-input λ ₀ , and within the value range set by λ to determine the The value of the covariate λ within the error tolerance. Step 2) is repeated, and the optimal control input sequence can be obtained after all calculations are completed.

7. The optimal control input sequence signal obtained by the solution is transmitted to the power execution control unit of the fuel cell vehicle. The optimal control input sequence signal obtained by the lower energy rolling optimization controller is transmitted to the power execution control unit of the fuel cell vehicle through the data bus. , acting on the fuel cell hybrid power system, so that the fuel cell vehicle distributes power according to the result calculated by the algorithm, and finally improves the fuel economy of the fuel cell vehicle.

8. Carry out experimental simulation to evaluate the energy-saving effect of the designed real-time energy optimization management system. Select the vehicle speed curve of the congested driving cycle (LA92) to verify the effectiveness of the designed system. The LA92 driving cycle curve is 1435 seconds long, and contains rich information on working conditions such as low speed and frequent start and stop.

According to the simulation results, it can be seen that the intelligent networked fuel cell vehicle real-time energy optimization management system proposed by the present invention has the following advantages:

(1) The upper layer of the designed real-time energy optimization management system for intelligent networked fuel cell vehicles

The SOC trajectory rolling optimization controller and the lower energy rolling optimization controller make full use of the network connection information in the macroscopic long-term domain and the vehicle speed preview information in the microscopic short-term domain to obtain the power battery corresponding to the optimal hydrogen consumption in the macroscopic prediction time domain The SOC trajectory provides sufficient reference information for the lower-level energy rolling optimization controller, and further taps the energy-saving potential of fuel cell vehicles.

FIG. 8 is a graph showing the vehicle speed of the selected congested driving cycle (LA92). Figure 11 is a comparison diagram of the reference optimal SOC ^* trajectory calculated by the upper-layer SOC trajectory rolling optimization controller under the congested driving cycle (LA92) and the actual SOC curve calculated by the lower-layer energy rolling optimization controller. It can be seen from the figure The tracking effect of SOC is good, and the SOC value is always within the constraint range. Figure 14 is the hydrogen consumption curve calculated by the real-time energy optimization management system of the IFC under the congested driving cycle (LA92), and Figure 15 is the real-time IFC fuel cell vehicle under the congested driving cycle (LA92). The comparison chart of the hydrogen consumption calculated by the energy optimization management system, the hydrogen consumption calculated by the rule-based energy management strategy, and the offline global optimal hydrogen consumption. It can be seen from the figure that the hydrogen consumption calculated by the real-time energy optimization management strategy of the connected fuel cell vehicle designed by the present invention is 169.93g, and the hydrogen consumption calculated by the rule-based energy management strategy is 199.32g, and the offline global optimal The theoretical optimal hydrogen consumption obtained by the strategy is 164.52g. By comparison, it can be seen that the designed system saves up to 17.3% of the hydrogen consumption compared with the rule-based energy management strategy, and the fuel economy is better than the rule-based energy management strategy. It is significantly improved, and is only 3.2% different from the theoretical optimal hydrogen consumption calculated by the offline global optimal strategy, which is close to the theoretical optimal hydrogen consumption, indicating that the designed system is very effective and can make full use of the rich network information. The hydrogen-saving potential of fuel cell vehicles greatly improves the fuel economy of vehicles.

(2) To ensure a reasonable and effective distribution of energy during the operation of the vehicle, the power of the fuel cell is always greater than 5kW, and the shutdown of the fuel cell stack is avoided during the operation of the vehicle. At the same time, the SOC of the power battery is always within the constraint range during the operation of the vehicle, which prevents the power battery from working in the low-efficiency range and prolongs the service life of the fuel cell and the power battery.

Figure 12 is the required power of the vehicle under the congested driving cycle (LA92) and the fuel cell output power and power battery output power calculated by the lower energy rolling optimization controller. Figure 9 is the congestion driving cycle (LA92) The optimal SOC ^* trajectory of the power battery calculated by the lower and upper SOC trajectory rolling optimization controller, Figure 11 shows the reference optimal SOC ^* trajectory and the lower energy calculated by the upper layer SOC trajectory rolling optimization controller under the congested driving cycle (LA92). The comparison chart of the actual SOC curve calculated by the rolling optimization controller. It can be seen from the figure that the SOC tracking effect is good, and its value is always between 0.45-0.52, which is the optimal efficiency range of the power battery, indicating that the fuel cell The energy distribution between the power battery and the vehicle transmission system is reasonable and effective, which further improves the endurance and service life of the fuel cell vehicle.

(3) The designed lower-level energy rolling optimization controller has a fast calculation rate and ensures the real-time performance of the system solution. The longest calculation time of the upper-layer SOC trajectory rolling optimization controller to solve the power battery SOC trajectory at all sampling times is 41.63 seconds, which is far less than 300 seconds, which can meet the update frequency of the upper-layer SOC trajectory rolling optimization controller every 300 seconds; Figure 13 shows the congestion The calculation time curve required by the energy rolling optimization controller to solve the output power of the fuel cell and the power battery at each sampling time under the operating condition driving cycle (LA92), it can be seen from the figure that the longest sampling time of the lower energy rolling optimization controller is the longest The calculation time is 0.052 seconds, far less than 1 second, which can meet the update frequency of sampling every 1 second, and ensure the real-time and rapidity of the system solution.

Combined with multi-scale network connection prediction information, the invention proposes a layered real-time energy rolling optimization control method for fuel cell vehicles, which effectively reduces the computational burden of real-time optimization of fuel cell vehicle energy management and greatly improves the energy efficiency of fuel cell vehicles.

The positive effects of the present invention are:

1. Designed a real-time energy optimization management system and system workflow for intelligent network-connected fuel cell vehicles. Using multi-scale network-connected information, the fuel economy was improved by 17.3% compared with the rule-based energy management strategy, and the fuel cell was further explored. The energy-saving space of the car;

2. Aiming at the prediction energy saving optimization problem of multi-power source fuel cell vehicles under multi-scale network information, a layered real-time energy rolling optimization control method for fuel cell vehicles is proposed, and the obtained fuel economy is close to the theoretical optimal economy; The maximum solution time for solving the fuel cell and power battery output power at all sampling times is only 0.052 seconds, which is far less than 1 second, which can meet the update frequency of sampling every 1 second. It not only effectively utilizes the preview information of different time scales to fully tap the hydrogen saving potential of fuel cell vehicles, but also can meet the real-time computing needs.

3. Using the network preview information in the macro and long time domain, the SOC trajectory rolling optimization controller is designed in the upper layer, and the cloud computing resources are used to solve the power battery SOC trajectory corresponding to the optimal fuel economy in this time domain. The trajectory information is sent to the lower-level energy rolling optimization controller to provide it with rich reference information and fully improve the energy-saving space of the fuel cell vehicle.

4. Using the microscopic short-time domain preview information, combined with the SOC trajectory reference information optimized by the upper layer rolling optimization, the energy rolling optimization controller is designed in the lower layer, and the on-board controller is used to quickly and efficiently solve the output power of the fuel cell and power battery.

Claims (1)

1. A real-time energy optimization management system for intelligent networked fuel cell vehicles, Step 1: Design a macroscopic long-term average traffic velocity trajectory prediction module; It is characterized by: Step 2: Design a microscopic short-term vehicle speed prediction module The neural network consists of three layers, namely the input layer, the hidden layer and the output layer; the input vector is defined as m(k), and the output vector is defined as

The structure of BPNN can be represented by a discrete model with weights and thresholds

where w ₁ is the weight between the input layer and the hidden layer, w ₂ is the weight between the hidden layer and the output layer, b ₁ is the threshold of the neurons in the hidden layer, and b ₂ is the threshold of the neurons in the output layer , m(k) represents the input historical speed sequence,

Represents the output predicted speed sequence, g(h) is the activation function from the hidden layer to the output layer, and its transfer function

Step 3: Establish a fuel cell vehicle power system model for energy optimal control 3.1 Establish the vehicle longitudinal driving dynamics model Fuel cell vehicle parameters: motor transmission efficiency η _{t_veh} (%), mass coefficient of rotating elements σ _veh (-), gravitational acceleration g (m/s ² ), air resistance coefficient C _{D_veh} (-), air density ρ _air (kg /m ³ ), vehicle mass m _veh (kg), windward area A _veh (m ² ), sliding resistance coefficient f(-), road gradient θ _road (-); demand power of the vehicle

where P _{veh_req} is the required power of the vehicle, f is the sliding resistance coefficient, η _{t_veh} is the motor transmission efficiency, σ _veh is the mass coefficient of the rotating element, m _veh is the vehicle mass, g is the acceleration of gravity, θ _road is the road slope, A _veh is the windward area of the car, ρ _air is the air density, C _{D_veh} is the air resistance coefficient,

is the derivative of the vehicle's speed V _veh with respect to time t; 3.2 Establish a fuel cell stack efficiency model Hydrogen consumption of fuel cells

Among them, P _{fc_req} is the output power of the fuel cell, η _{fc_st} is the working efficiency of the fuel cell,

is the lower calorific value of hydrogen; 3.3 Establish a power battery SOC model P _{batt_req} =P _{veh_req} -P _{fc_req} (5) Among them, P _{batt_req} is the output power of the power battery; The SOC dynamic equation of the power battery is:

Among them, V _{oc_batt} is the open circuit voltage of the power battery, R _{int_batt} is the internal resistance of the power battery, Q _batt is the total power of the power battery,

is the derivative of the power battery state of charge SOC; Step 4: Establish the energy optimal management problem 4.1 The state equation of the output power P _{fc_req} of the fuel cell is:

The above formula can be obtained by the formula

express; Minimize the hydrogen consumption of the system in the predicted time domain [t ₀ ,t _f ]:

Among them, J is the total hydrogen consumption in the prediction time domain under the condition that the system terminal constraints are satisfied, t ₀ is the start time of the prediction time domain, t _f is the end time of the prediction time domain, u is the control input variable, and U is the control input variable. The set of values for the input variable,

The hydrogen consumption of the system at time t is a function related to the control input u(t) variable at time t. The control input variable takes u=P _{fc_req} , the state variable takes x=SOC, and φ(x(t _f )) is Terminal constraints on state variables; 4.2 The constraints that are satisfied are as follows: (1) The output power constraints of the fuel cell need to be met: P _{fc_low ≤P fc_req} (t) _{≤P fc_up} (10) Among them, P _{fc_low} is the minimum output power of the fuel cell, P _{fc_up} is the maximum output power of the fuel cell, and P _{fc_req} (t) is the output power of the fuel cell at time t; (2) The dynamic equation and state constraints of the power battery SOC need to be satisfied:

Among them, SOC _begin is the SOC value of the power battery at the initial time, SOC _low is the minimum value of the SOC of the power battery, SOC _up is the maximum value of the SOC of the power battery, SOC(t) represents the value of the SOC of the power battery at time t, SOC(t ₀ ) represents the value of the SOC of the power battery at the initial moment, and SOC(t _f ) represents the value of the SOC of the power battery at the terminal moment; (3) The power constraints of the power battery need to be met P _{batt_low ≤P batt_req} (t) _{≤P batt_up} (12) Among them, P _{batt_low} is the maximum charging power of the power battery, P _{batt_up} is the maximum discharging power of the power battery, and P _{batt_req} (t) is the output power of the power battery at time t; (4) It is necessary to meet the demand power when the car is running P _{veh_req} (t)=P _{batt_req} (t)+P _{fc_req} (t) (13) Among them, P _{veh_req} (t) is the required power of the car at time t, and P _{batt_req} (t) is the output power of the power battery at time t; Step 5: Using the long-time domain preview information, design the upper-layer SOC trajectory rolling optimization controller 5.1 Upper-layer SOC trajectory rolling optimization controller optimization problem Discrete the prediction time domain [t _0,m ,t _f,m ] under the time scale into N _m equal parts, where t _0,m is the start time of the prediction time domain, and t _f,m is the prediction time The termination time of the time domain, the discrete time is denoted as k∈{1,2,...,N _m +1}, and the optimization objective is obtained:

Among them, J is the total hydrogen consumption of the system at all sampling times under the condition of terminal constraints, φ(x(N _m +1)) is the terminal constraints of the state variables,

The representative hydrogen consumption is a function related to the control input u(k) at time k, Δt is the sampling time interval between two adjacent vehicle speed information, and the control variable u(k) is the output power of the fuel cell at time k P _{fc_req_m} ( k); The specific constraints that are met are: (1) Satisfy the output power constraints of the fuel cell: P _{fc_low ≤P fc_req_m} (k) _{≤P fc_up} (15) (2) Satisfy the dynamic equation and state constraints of the power battery SOC in this time domain:

Among them, SOC _m (k+1) is the value of the SOC of the power battery at the next moment in the time domain obtained after the control input of the power battery SOC _m (k) at time k, that is, the value of the SOC of the power battery at time k+1 , V _{oc_batt_m} (k) is the open circuit voltage of the power battery at time k in this time domain, R _{int_batt_m} (k) is the internal resistance of the power battery at time k in this time domain, P _{veh_req_m} (k) is the required power of the car at time k , P _{fc_req_m} (k) is the output power of the fuel cell at time k in this time domain, SOC _m (k) is the state of charge value of the power battery at time k in this time domain, SOC _m (1) is the power battery in this time domain The value of the battery SOC at the initial time, SOC _m (N _m +1) is the value of the power battery SOC terminal time in this time domain; (3) Satisfy the output power constraints of the power battery: P _{batt_low ≤P batt_req_m} (k) _{≤P batt_up} (17) Among them, P _{batt_req_m} (k) is the output power of the power battery at time k in this time domain; (4) To meet the demand power when the car is running P _{veh_req_m} (k)=P _{fc_req_m} (k)+P _{batt_req_m} (k) (18); 5.2 Meshing about system states and control variables The state variable power battery is divided into 81 state grids; the output power of the fuel cell increases by 81 control variable grids from the beginning; 5.3 Calculate the cost of consideration Under the action of the control variable u(k), the state variable x(k) will be calculated by the state transition equation to obtain a new state variable x(k+1). From time 1, different control variable grids act on the state The state variable grid at the next moment will be obtained on the variable grid, resulting in the corresponding cost J(k), and the new control variable grid will act on the state variable grid at this moment to generate the corresponding cost at the next moment. The cost J(k+1) is calculated until the entire driving cycle is completed, and the resulting cost can be calculated by the formula

After the calculation is obtained, the cost generated by each iteration from the forward to the backward is stored in the grid; 5.4 Determining the optimal decision Determine the value of the state variable at the terminal moment k=N _m +1, that is, x(N _m +1), corresponding to the initial objective function J(N _m +1)=0, then from the last moment of the terminal moment:

Among them, J ^* (k) represents the minimum value of hydrogen consumption when the system state variable at the kth time is x(k); L(x(k), u(k)) represents the kth time, the system is in the state variable The hydrogen consumption generated by x(k) after the action of the control input u(k), that is, the state transition cost, J ^* (k+1) is the hydrogen consumption when the system state variable is x(k+1) at the previous moment The minimum value, select the state variable corresponding to the minimum value of the cost function from each moment, and then the optimal state variable sequence {x ^* (1),x ^* (2),...,x ^* (k )}, that is, the optimal power battery SOC sequence SOC ^* ; Step 6: Use the short-time domain preview information to design the lower-level energy rolling optimization controller 6.1 Receive the SOC ^* trajectory sequence in the predicted time domain at the current sampling time, and read the SOC value of the current power battery; 6.2 The lower energy rolling optimization controller optimization problem The vehicle speed preview information in the microscopic short-time domain [t _0,n ,t _f,n ] is discretized into N _n equal parts, and the discrete time is denoted as s∈{1,2,...,N _n +1}, where t _0,n is the start time of the prediction time domain, t _f,n is the end time of the prediction time domain, and the optimization objective function is obtained:

Among them, _{ud is the control input in this time domain, U d is} the set of control input values within the range of the control input in this time domain, and I is the SOC of the power battery at all sampling times of the system under the condition that the system terminal constraints are satisfied The sum of the square of the difference with SOC ^* and the sum of hydrogen consumption, SOC _n (s) is the value of the power battery SOC at s time in this time domain, φ(x _d (N _n +1)) is the terminal constraint of the state variable , _ud (s) is the value of the control input at time s in this time domain,

It represents that the hydrogen consumption is a function related to the control input _ud (s), the control variable is selected as _ud =P _{fc_req_n} (s), and the state variable is selected as x _d =SOC _n (s). The specific constraints are: (1) Satisfy the output power constraints of the fuel cell: P _{fc_low} <P _{fc_req_n} (s) < P _{fc_up} (21) (2) Satisfy the dynamic equation and state constraints of the power battery SOC:

Among them, SOC _n (s+1) is the value of the SOC of the power battery at the next moment in the time domain obtained by the SOC _n (s) of the power battery at the time s in the time domain, that is, the power battery at the time s+1. The value of SOC, V _{oc_batt_n} (s) is the open circuit voltage of the power battery at time s in this time domain, P _{veh_req_n} (s) is the required power of the car at time s in this time domain, and P _{fc_req_n} (s) is s in this time domain The output power of the fuel cell at the time, R _{int_batt_n} (s) is the internal resistance of the power battery at the time s in the time domain, SOC _n (1) is the SOC value of the power battery at the initial time in the time domain, SOC _n (N _n +1 ) is the value of the power battery SOC at the terminal moment in this time domain; (3) Satisfy the output power constraints of the power battery: P _{batt_low ≤P batt_req_n} (s) _{≤P batt_up} (23) (4) To meet the demand power when the car is running: P _{veh_req_n} (s)=P _{fc_req_n} (s)+P _{batt_req_n} (s) (24) 6.3 Constructing the Hamiltonian

Among them, H(x _d (s), u _d (s), λ(s), s) represents the value x _d (s) of the Hamiltonian function and state variables at time s, and the value of control input at time s _{d d} (s), the value λ(s) of the co-state variable at time s is related to the current time s, and the necessary conditions to be satisfied for optimization are as follows:

λ(s+1)=λ(s)+Δλ(s)·Δt, (26) Among them, Δλ(s) is the difference between the co-state variables at two adjacent moments,

Represents the value of the partial derivative of the Hamiltonian function at time s to the SOC of the power battery, λ(s+1) is the value of the co-state variable at the next time in the time domain obtained by calculating the co-state variable λ(s) at time s , that is, the value of the covariate at time s+1; at the same time, the optimal control input

It is necessary to ensure the minimum Hamiltonian function at each sampling moment, that is, the following formula needs to be satisfied:

in,

is the optimal value of the state variable at time s,

is the optimal value of the control input at time s, λ ^* (s) is the optimal value of the covariate variable at time s,

Represents the optimal Hamiltonian function and the optimal value of the state variable at time s

The optimal value of the control input at time s

The optimal value λ ^* (s) of the covariate at time s is related to the current time s,

Represents the optimal value of the Hamiltonian function and the state variable at time s

The value _ud (s) of the control input at time s, and the optimal value λ ^* (s) of the co-state variable at time s are related to the current time s; 6.4 Solving the Optimal Control Input Sequence (1) Set the state variable SOC ₀ at the initial moment, and calculate the initial value λ ₀ of the co-state variable at the initial moment by the bisection method; (2) The control input when the Hamiltonian function value is the smallest is the optimal control input

U _d (s)=[u _{d_low} (s):Δu _d (s):u _{d_up} (s)], (28) Among them, _Δud (s) is the difference between two adjacent equal parts of the control input at time s in the time domain, _{ud_up} (s) is the maximum value of the control input within the constraint range at time s, and _{ud_low} (s) is The minimum value of the control input within the constraint range, U _d (s) is the set of values of the control input at time s in this time domain; (3) According to the optimal control input action

At the result of the state transition equation, calculate the value of the state variable SOC and the value of the co-state variable λ at the next sampling time, and repeat step 2) until the last sampling time; (4) Judging the error value of the last terminal boundary of the SOC value, if the error is within the set range, end the calculation, otherwise it is necessary to re-input λ ₀ , and within the value range set by λ to determine the The value of the co-state variable λ within the allowable error range, repeat step 2), and the optimal control input sequence can be obtained after all calculations are completed; Step 7: Transmit the obtained control input sequence signal to the power execution control unit of the fuel cell vehicle.

Images